What type of program for study
Six Sigma – iSixSigma › Forums › Old Forums › General › What type of program for study
 This topic has 12 replies, 6 voices, and was last updated 14 years, 7 months ago by Robert Butler.

AuthorPosts

April 29, 2007 at 6:31 am #46844
I am a high school junior doing a study for Intel. My research involves comparing nutrigenomic based supplementation for methylation, measuring improvements in 2 lab values over time, and then measuring behavioral improvements over the same time period. Trying to prove that an improvement in lab values = improvement in behavior.
N=will be 30. Have no idea what type of analysis I would do for this,
Trying to make this as powerful as I can, as I can not do a control study so need it to be strong (I know it is, just have to prove it)
Thanks for any suggestions/help
Meg0April 29, 2007 at 7:04 am #155420
Allthingsidiot OParticipant@AllthingsidiotO Include @AllthingsidiotO in your post and this person will
be notified via email.I may suggest to apply the “paired tTest”,using the formulas:
t=d/Sd/SRn & t=d(bar)mu/Sd/SRn (SR stands for square root).
You may state H0: Mu1=Mu2
H1:Mu1 n0t equal Mu2 (using the first formula above to calculate the t ),then search for t(0.025,29) in the concerned ttable,if you receive a higher value than the calculated one then you conclude :fail to reject,if below then you would reject H0 and accept H1.
Good Luck0April 30, 2007 at 10:50 am #155468
accringtonParticipant@accrington Include @accrington in your post and this person will
be notified via email.I don’t quite understand what your question is. When you say you are measuring behavioural improvement, are you talking about people? If so, does N = 30 refer to the number of subjects in the study?
0April 30, 2007 at 7:13 pm #155513Yes 30 subjects measuring two lab values from pre/then post.So two values measured twice (start/end) Then a behvaioral assessment pre/then post(will be at the same time as the above values are obtained), the behavior assessment measures 4 categories, but it will already be scored and a total then a subcore for each category. So a total score and 4 individual values pre and then again post on each 30 subjects.
0May 1, 2007 at 12:35 pm #155538
accringtonParticipant@accrington Include @accrington in your post and this person will
be notified via email.To clarify. Are there two different values that you are measuring in the lab, and are they continuous variables?
For the beahvioural assessment, there are four categories, plus the total of the four categories. Are these continuous variables?0May 2, 2007 at 4:55 pm #155615If you mean continuous, blood value MMA will be measured pre/blood value MMA will be measured post. ) Blood value B will be measured pre treatment then /Blood value B will be measured post treatment . Same value measured twice, but two different markers. (forgive my non statistical terms, only high school junior) All 30 subjects will be receiving the same substance.
Then I would like to correlate the improvements in the 2 different markers in the blood to the behvioral assessment, which has a total score, but then will have a sub score measuring 4 different domains. The 4 subcatgories are:
1. speech
2. sociability
3. sensory cognitive avareness
4. Physical behavior
Same questions pre test post test. Each sub category gives a numerical score, then all added for a total score.
I don’t have the ability to do a placebo/double blind study with this population.
Thank you for your help, and I hope this is more clearly explained.0May 2, 2007 at 8:06 pm #155628This is a test with many moving parts so it is important to set up your experiment correctly to minimize bias as much as possible.
You have a chain of hypotheses that must be tested along the way. First, you hypothesize that a given supplement will have a statistically significant effect on blood chemical A. Second, you hypothesize that the same given supplement will have a statistically significant effect on blood chemical B. You can test both of those hypotheses using a paired ttest as was previously suggested. You must run two seperate paired ttests; one on blood chemical A before/after and one on blood chemical B before/after.
After you have determined significance for each blood chemical of interest, you can do a correlation study to see if the change in blood chemicals correlate to the behaviors you outlined. I would expect that there may be a mild correlation, but I would expect a very large error component due to the small sample size and to the uncontrollable factors in the study.
Good luck with your study, it sounds interesting.
0May 3, 2007 at 4:12 am #155646Thank you for explaining how to set it up. Knowing my sample size is small, but one more question, what would be a statistically strong p value for a study like this, I know many errors, but just wondering?
Thanks again0May 3, 2007 at 12:01 pm #155651
Mike ArcherParticipant@MikeArcher Include @MikeArcher in your post and this person will
be notified via email.Hello. I’m new in training, but I did learn that a 99% confidence interval is needed for any hypothisis test related to medicine or health. That equates to a pvalue of 0.01 or less to reject the null.
Mike0May 3, 2007 at 6:08 pm #155671Mike,
THANKS! So far what I am getting is 0.01 so far so good. Phew! Three years on this and thought it would be no good. Would that pvalue give my hypothesis more power? Even with such a low number of participants?0May 3, 2007 at 7:04 pm #155677
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.I don’t know the sources of the training material that would suggest that one needs a p value of .01 or less for the medical field. I’ve been working in the medical field for quite some time and the cut point in all of the work I’ve done has been p < .05.
As for the question about p value giving the hypothesis more power – this doesn’t make sense. The power of a test and the degree of significance of the results are not dependent on one another. For the kind of work you are doing (we would call it a pilot study) the usual practice is to focus on alpha (look for results with p <.05) and recognize that, more likely than not, the study will be very underpowered.
The reason for doing an underpowered study is simple – a pilot study’s primary focus is on trying to see if there is anything worth investigating. Running an initial study with hundreds of mice/enrolled patients/volunteers etc. only to declare, with a power of .8 or .9, that there wasn’t anything of value is a waste of time and money and a sure fire way to guarantee a short career path in the world of medicine.
0May 4, 2007 at 2:29 am #155693So in other words, this would be considered a pilot study, and if the alpha value is p<0.05 one would state the studies limitations due to size of sample, but that it might be worth larger study? Is that correct?
0May 4, 2007 at 12:09 pm #155702
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.The short answer to your question is – yes. When writing up our findings we report them in the following way:
1. A simple table with the population demographics – that is a bean count for things such as age, gender, race, prior relevant medication, etc.
2. If we have an effect with P < .05 we will report this and we will also use the sample statistics to give the reader an estimate of the power associated with our study.
3. If the effect has a P < .05 and if the difference is clinically meaningful we will comment on its ramifications in the discussion section and we will also highlight the study shortcomings (size of sample, bias of sample, clinical issues not addressed etc.) and make recommendations concerning possible future work.
Before going much further in your work I’d recommend you get a copy of An Introduction to Medical Statistics 3rd Edition by Bland and give it a reading. This is the book we use when introducing our residents/fellows to issues surrounding the use and reporting of statistics in the medical arena.0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.